Is there Dynamic Inefficiency when the rate of growth of GDP is greater than the safe real interest rate

but lower than the expected average return on capital ? *

The question is technical but has great policy relevance. Another way to phrase it is to ask if it is
plausible that, in the absence of Ricardian equivalence, but under the assumption of full employment,
an increase in public debt would cause higher balanced growth welfare. I think that useful light on this
question can be cast by a simple model with strong assumptions.

Throughout I will maintain some extreme simplifying assumptions. The first is that employment is
exogenous: there is full employment & labor supply is exogenous. This note has nothing to do with
Keynesian arguments about the effect of public debt if full employment is not guaranteed by
assumption.

The second is that the economy consists of a state plus overlapping generations of agents who work
when they are young and live off of their capital when they are old (standard Diamond model with
productive capital). This means that there is not Ricardian equivalence and the timing of taxation
affects the economy even given the public sector budget constraint. It also means that, in theory, the
economy can be dynamically inefficient so that a lower constant capital to effective labor ratio causes a
higher consumption to effective labor ratio.

The state will do only one thing. It issues safe one period inflation protected securities which are sued
in period t and worth one unit of consumption good when they mature in period t+1. These payments
are financed by the issuance of new securities and possibly by lump sum taxes. The state may also
make lump sum transfer payments to the young or to the old.

For simplicity only assume that the population (and labor supply) are constant and normalized to 1.
The effective labor supply grows only because of exogenous labor augmenting technological progress.
This just simplifies notation compared to a standard model with both population growth and
technological progress.

It is assumed that each old person (of a continuum of measure 1 indexed by i) operates his or her own
firm (these firms are liquidated after one period). Each firm has an idiosyncratic firm specific
technology.

Output by firm i at time t is given by a standard Cobb Douglas production function.

α 1α
Y it =K it ( Ait Lit ) (1)
Where

A it = A t (1+ǫ it )1 /(1α ) (2)

Each firm has an idiosyncratic firm specific technology shock. ǫit has mean zero and variance (σ)2 .

So
 A 1α 1α
it = A t (1+ǫit ) (3)

This shock is not verifiable (that is output is not verifiable) so each owner must bear all of the risk. It
can't be born by outside shareholders or by the employee.

At grows geometrically

A t =A t 1 (1+ g) (4)

Each old person i chooses Lit before ǫit is realized so Lit = 1 * and

α 1α
Y it =K it A it (5)

I thank Nick Rowe for explaining to me that the assumption that employment is chosen before the
shock is realized is a better motivation for the the result that all firms have the same number of
employees than the odd assumption I originally made that each old person has the capacity to supervise
no more than one young person.

In period t the state sells bonds for B t=bAt units of consumption good. In period t+1 the bonds are
redeemed for bA t (1+r ts+1 ) where r st is the safe rate of interest. bA t+1 =(1+ g) bAt and the state
distributes a pension of bA t ( gr st +1) to the old.

Young agents divide their Wt income into consumption C ty , investment in their new firm K t +1
and bond purchases bA t . They are identical and make identical choices so the subscript i is
supressed.

Cty + bAt +K t+ 1=W t (6)

Young workers supply labor inelastically. The old decide how many young workers to hire (in
equilibrium each hires one) promising pay W_t. Then uncertainty is resolved, production occurs, the
old pay w_t and consume

Cito =bA t + K t +K αt A 1α

it W t (7)

There is no completely standard notation for the time subscript on consumption of the old. Here the t
subscript refers to the time when the consumption occurs. This notational problem will not create
difficulties because I will consider only balanced growth paths.

The expected return on capital paid in period t is (1α) k αt 1 which must be greater than r st .
If
(1α) k tα1> g (8)

Then capital income is greater than investment, that is, total consumption is greater than total labor
income. In models without risk, this is the condition for dynamic efficiency. It is however, consistent
with dynamic inefficiency in this model. I assume that the inequality holds.
Assume agent i born in period t maximizes the expected value of

ln(C ity )+β ln (C oit+1 ) (9)

This is the mathematically easiest of standard utility functions chosen for convenience.

The present value of the wage plus the discounted pension is

bA t (gr st +1)
permanent income=W t + (10)
1+r ts+1

Even for the case of logarithmic utility, there isn't a closed form solution for Cty as a function of w_t
bA_t and the safe interest rate r st +1 . The only simple case occurs when r st +1 = g. then no matter
what the risky return on capital is, a constant fraction of the wage

s y Wt
if g=r t+ 1 thenC t = (11)
1+β

In this special case, government bonds crowd out private investment one for one with

β Wt
K t +1= bA t (12)
1+β

If r st +1 < g then government bonds crowd out private investment more than one for one as the
expected pension payment causes higher consumption when young.

In this model the labor demand problem is a bit tricky, because the old don't know A it when they
chose to hire 1 young worker at wage W_t. At time t the old choose Lit (which will be equal to one in
equilibrium) to maximize the expected value of utility

E[ln ( bAt + K t + K αt ( A it Lit )1αW t Lit )] (13)

Equation (14) is a second order Taylor series approximation to (13)

with which implies the first order condition at Lit = 1

 α 1α
(1α) K t Ait W t
E =0 (14)
K t +bA t + K αt A1α
it W t

Now define k = K/A and w = W/A and note that A 1α

it / A1α
t =1+ǫit

So the first order condition becomes

(1α) k αt (1+ǫit )wt

E =0 (15)
k t +b +k αt (1+ ǫit )wt
α
Which implies w t <(1α) k t

The wage is lower than it would be in the absence of risk, because the old's marginal utility of
consumption is higher when A_t is low. Another way of putting this is that the risk reduces labor
demand, because higher employment causes the employer to bear more risk.

For small enough σ2 one can use a second order Taylor series approximation to the utility function
to obtain an approximate first order condition at at Lit = 1

(1α) k αt wt 2(1α)k αt (k t +b+ k αt wt ) σ2

α  ≈0 (16)
k t + b+k t w t 2(k t + b+k tαwt )2

Which immediately simplifies to

w t≈(1α)k αt (1σ2 ) (17)

Finally assume balanced growth, so the lower case variables are constant. If rs = g balanced growth
occurs iff

K t +1 βw
k (1+ g)= =k + b (18)
At 1+β

βw β(1α) k α
Recall that b<k + b (19)
1+β 1+β

If rs<g balanced growth k is reduced due to the effect of pensions on saving (this effect increases in b).

The welfare of an agent born at time t is

 w α
U t =ln( )+β E[ ln( b+k + k (1+ǫ it )w)]+( t +1) ln (1+ g) (20)
(1+β)(1+ g)
An increase in b has three effects in equilibrium. It causes a shift of saving from risky capital to safe
bonds. For constant expected return on capital, If rs and w, agents are indifferent about this shift. If If
rs<g, it causes an increase in pension payments which unambiguously cause higher welfare and it
causes a reduction of w both by causing reduced k and by causing reduced expected consumption and
increased absolute risk aversion of the old. The reduction in w implies a transfer from the young to the
old with each agent receiving 1+g units of consumption when old for each unit of consumption lost
when young. Since w is not stochastic, this transfer involves no risk. If rs<g this transfer causes
higher welfare for each generation. Agents are indifferent about consuming a bit less when young,
saving a bit more and receiving a safe return of rs g so they strictly prefer the effect of reduced wages
which are equivalent to making them save and receive a safe return of g.

This means that, if the safe rate of interest is less than the rate of growth, the economy is dynamically
inefficient. This is the case even though total capital income is greater than investment, that is, total
consumption is greater than labor income.

The usual simple national income and product accounts based formula for determining whether an
economy is dynamically efficient due to Abel, Mankiw, Summers, and Zeckhauser does not work in
this case.

More conventional models.

The strong conclusion above follows partly from the strong, and unusual, assumption that uncertainty is
resolved after firms chose how many workers to hire. I don't have any particular sense that the
standard assumption that employment is chosen after uncertainly is resolved is more reasonable, but I
should note that it can make a big difference.

It is still possible, with the extreme one firm each assumption, to get the results exactly as above. I
think the best way to do this is to take very literally the assumption of one firm per old person and
assume each old person can hire 0, 1 or 2 young workers. Then have the highest possible productivity
of the second employee as low is equal to the lowest possible productivity of the first employee. This
means that each old person will choose to hire exactly one employee and the wage will be equal to that
productivity.

The fact that employment must be an integer implies that, for the firm with the lowest productivity, the
worker gets all of the output and profits are zero.

Everything else works out just as in the model above assuming output of a firm with one employee is
equal to
α 1α
Y it =K it ( Ait ) (21)

And the minimum value of epsilon is given by inequality 22

ǫ≥(1α)(1σ 2)
Again the result is that public debt crowds out investment and that this causes higher welfare so long as
the rate of average technological progress n is greater than the safe interest rate.

However, the assumption of one firm per worker is unrealistic. It might be accepted if it is clearly just
a way to model financial frictions, but it is unlikely that anyone will take seriously this model in which
the assumption is taken seriously. Rather the standard assumption is that there labor is continuous and
there are perfect returns to scale. This may imply workers moonlighting – working for more than one
firm, but it is so standard that it is what I assumed in the first model I discussed without even noticing
the strangeness of the assumption.

Model III, an actually conventional model with continuous employment and technology resolved
before employment is chosen behaves quite differently than the eccentric models above. Firms all pay
the same wage wt so

Lit =(1α)1/ α K A it(1α)/ α W 1/

t
α
(22)

Average employment equals one so

W t =(1α) K t α A t1α ( E((1+ǫit )

(1α)/α α
)) (23)

Which is standard except for the extraordinarily ugly term in ǫ

Again public debt will crowd out investment and cause lower wages. Again this amounts to a transfer
from the young to the old. However, in this case, the benefit of the reduced wage is greater to the old
agents who are already lucky because they have a high Ait. This makes the effect of reduced wage a
risky transfer just as investment itself is risky and, in fact, with the exact same pattern of risk and
return.

Output of firm i is

Y it =K t ((1α) A it )(1α )/α W (α1)/α

t (24)

And, given the conventional labor demand with Cobb-Douglas production, the share of capital is α
so the consumption of old person i is

Cito =bA t +K t ( 1+ α( K t ((1α) A it )(1α )/α W (α1)/α

t )) (7)

This means that the benefit from reduced wages is proportional to total profits. The reduction in wages
multiplies the return to investment by a constant greater than one. The distribution around the means
of the gain in consumption when old due to an additional unit of saving and due to a unit reduction in
wages is identical.

However the average gain to the old due to the reduction in wages is not equal to the average return on
capital. Instead, agents gain 1+g units of consumption on average when old in exchange for each unit
of reduced wages when young. This means that the wage reduction due to crowding out effect of
public debt is like a risky investment with return g. This reduces welfare if the risky return (rr) that is
the expected return on capital, is greater than g.

In the more conventional model, public debt has two contrasting effects when the rate of growth is
higher than the safe interest rate and lower than the expected risky return – the pension increases
welfare but the reduction in wages reduces welfare.

It is clear that if the rate of growth is equal to the expected return on capital, then public debt causes
higher welfare. This means that the standard Abel, Mankiw, Summers, and Zeckhauser calculation is
misleading in this case as well.

Here is a very very rough back of the envelope calculation for the USA of the relative magnitudes of
the gain to the public sector of a higher debt to GDP ratio and the loss to private agents due to crowding
out and reduced wages. First define lower case variables as upper case variables divided by At so
kt = Kt/At etc,

If debt = bAt the state can transfer (g- rs) bAt each period. For my convenience, I will assume that part
is given to the young and part is given to the old so the transfer does not affect the amount agents
choose to save (if all were given to the old the transfer would lower saving and if all were given to the
young it would cause higher saving). This causes an increase in steady state welfare larger than the
benefit of transfers only to the young -- there is a term in (g- rs) which I will ignore.

K t+1 +bA t β βw t
Logarithmic utility implies a constant saving rate = so (1+g)k t+1 = b .
Wt 1+β (1+β)
∂w t α w t ∂ kt kt
Cobb Douglas production function implies = so = so in balanced growth
∂ kt kt ∂w t α w t
with constant k

dk β dw (1+g)k dw
(1+g) = 1= (8)
db (1+β) db α w db

(1+g) k β
At b = 0 =
w 1+β

so at b=0

dw ( 1+β)α
= (9)
db β(1α)

This means that a small positive debt to technology ratio b has a first order effect on welfare the same
as a windfall of the order of

(1+β) α
[( gr s)+(gr r ) ]bAt (10)
β(1α)
The sign depends on the endogenous safe and risky interest rates and the parameters α and β. The
assumption α = 1/3 is conventional. The unit of time is one generation, so β is small and the interest
rates are very high. Consider a GDP growth rate of 2.5% per year. If a generation is roughly 28 years
long, 1+g is roughly 2 . 1+rr is very roughly on the order of 3 (corresponding to a risky rate of roughly
4% per year) to 9 (corresponding to a risky rate of roughly 7% per year) .

r r =α k 1α So

w 1α r
=(1α) k =(1α)r /α (11)
k
and
β(1α) r
(1+g) k =2 k= r k (12)
(1+β) α
r
dw r
So =
db 2

dw
For the low risky rate of roughly 4% per year or 200% per generation, this implies that ≃1 and
db
the introduction of a small public debt causes increased welfare if g-rs > rr – g that is if rs < 0% per
year. For this unusually low estimate of the risky return, the model implies dynamic inefficiency only
if safe real interest rates are negative.

The appropriate safe interest rate is the 30 year inflation protected (TIPS) yield which is currently
Yielding 0.83% per year, so the model with employment determined after uncertainty is resolved
suggests dynamic efficiency.

Finally, an informal discussion of a genuinely conventional model. There are standard real business
cycle models with risky returns on capital in which the risk is aggregate risk. These models have
perfectly efficient financial markets. If such assumptions about production are introduced to an OLG
model, the model more difficult to analyze because wages, expected returns on capital and aggregate
capital are stochastic. However, it is clear that the benefit to the old due to reduced wages is risky.
This means that the analysis of the model III should apply to expected values and balanced growth in
the conventional OLG model with aggregate risk.

I am not at all willing to continue to try to work out the model (I have tried and not gotten much of
anywhere) but I think it is likely that a persistently negative safe real rate of interest is required for
dynamic inefficiency.

Both the conventional model with aggregate shocks and the conventional model with idiosyncratic
shocks followed by flexible choice of employment include standard but very strong assumptions. First,
as stressed above, the models assume full employment so increased consumption due to public debt
crowds out investment one for one. If there is any effect of aggregate demand on output at all, the case
for increased debt is stronger. Second it is assumed in the conventional models that firms can adjust
employment without any cost. This means that, in the models, labor income is always proportional to
capital income. In historical data, the share of labor is markedly countercyclical. This suggests that a
valid model would assume partial adjustment of employment to shocks which would imply that the
critical safe interest rate which implies dynamic inefficiency is positive but smaller than the growth
rate. Finally, all the models assume perfect competition in product markets. If firms have market
power, then rents are treated as capital income in national accounts and the true marginal product of
capital would be less than the ratio of capital income to capital. The risky rate of interest is the
expected marginal product of capital. If it is significantly over estimated because of imperfect
competition the US economy may be dynamically inefficient even though safe real interest rates are
positive.

The fact that interest rates are close to the critical values calculated with models based on extreme
assumptions which tend to imply dynamic efficiency convinces me that it is reasonably likely that
increased public debt would cause increased welfare.

* I want to thank Nick Rowe for a very useful comment.